Install the application that turns your Windows or Linux machine into a SocketJack AI host. Jack LLM Workstation connects local models, tool permissions, project files, terminals, SQL, and remote browser sessions without hiding the work from the computer that owns it.
This is the page users land in after choosing a SocketJack host: a live chat workspace with server routing, token status, model modes, GPU origin controls, multi-GPU pooling, tool progress, and a prompt composer in one browser surface.
The Windows Jack LLM Workstation installer includes the SocketJack VSIX installer. Leave the Visual Studio extension option checked to install the Copilot server picker during setup, or clear it when you only want the workstation service.
After Visual Studio restarts, open Extensions > SocketJack > SocketJack Copilot Servers, select a server, choose an eligible tools-capable model, test the route, and click Configure.
A workstation is more than a chat box. It is the local control surface for models, tools, project context, permissions, server health, and browser-steered sessions.
SocketJack.LlmRuntime coordinates local model providers, runtime health, generation requests, media tools, and model availability from the host machine.
SocketJack.Net handles the HTTP, WebSocket, proxy, relay, and server-list plumbing that makes a private workstation reachable from SocketJack sessions.
File, terminal, SQL, internet, and automation capabilities stay behind workstation-side permissions so hosts expose only what they choose.
Watch model status, usage, token settings, runtime reports, and workstation health before publishing the machine to the master list.
Attach project files, generated artifacts, and tool outputs to a session so useful work remains tied to the same local workspace.
Start at SocketJack.com, route prompts to your workstation, stream results back, and keep control without moving every workflow to a remote service.
The application, AI technology, and networking stack each have a clear job, which makes the platform easier to explain and easier to evolve.
Start in the main Jack LLM Workstation application, add the Visual Studio extension when you want Copilot routing, then publish or select a server from SocketJack.
The main application is the control room for proxy status, LlmRuntime, JackONNX, health metrics, service diagnostics, trust settings, models, and server management.
The Models tab shows running and downloaded inventory, benchmarks, backend selection, GPU layers, placement, target GPUs, context length, batch size, KV-cache offload, fallback, and credentials for each loadable model.
Recommended path for launching quickly with automatic server selection.
Optional advanced path when you want to pick a specific server yourself.
Jack LLM Workstation lets each model carry its own load profile, so small models can stay responsive while larger GGUF or media models get deliberate hardware placement.
Choose CUDA 12, DirectML, CPU, or backend fallback behavior per model. Use GPU Layers to control how much of a model is offloaded, including full auto offload when the backend supports it.
Target a specific GPU such as cuda:0, leave targets empty for automatic placement, or use hybrid and parallel modes when a model should span local devices or network nodes.
Set maximum GPU load and VRAM percentages so the workstation keeps room for the desktop, multiple running models, or rented sessions. Idle unload keeps heavyweight models from squatting on memory.
Adjust context length, eval batch, flash attention, and KV-cache offload to trade memory use for response speed, longer prompts, or steadier multi-session throughput.